Performance

Performance behavior

The main risk is doing too much work on rapidly mutating AI pages. v0.14 reduces that risk while keeping capture and health responsive.

Content-path budgets

• No content-script network requests.
• Mutation work is coalesced through requestIdleCallback (timeout fallback 120ms).
• Pending roots are nested-root deduplicated and capped at 40; overload collapses to one main/document scan.
• Capture scans only added/changed roots, not the whole document on every mutation.
• Tool evidence uses targeted selectors, scans at most the last 220 candidates, and is cached until relevant mutations dirty it.
• Semantic events are batched (up to 120 per message) rather than sent one by one.
• Hidden tabs disconnect the mutation observer and slow health/status timers to 30 seconds.
• Default Execution Pulse cadence is 2.5 seconds while active and 12 seconds while idle.
• Service-worker health context is cached for five seconds.

Starfield budget

The atmospheric background is one pointer-free pseudo-element with sparse radial gradients and a single translate3d animation over 95–115 seconds. It does not animate particles individually. Reduced-motion disables animation. Devices reporting four or fewer logical cores or 4 GB or less device memory use a static layer; effects can be removed entirely.

Pressure relief

The Performance Engine watches long tasks in a rolling window. When adaptive motion relief is enabled and measured pressure is high, it shortens only decorative aria-hidden animation. It never hides, virtualizes, or deletes provider conversation content.

Request governance

Background provider reads are serialized per provider, enforce minimum intervals, use freshness windows/conditional headers, honor Retry-After, and back off on 429/errors. Global connection refresh does not fan out provider HTTP probes; explicit provider checks do.

Verification

content_smoke.py exercises 320 mounted turns and asserts full retention, bounded runtime messaging, and zero page errors. live_health_smoke.py verifies progress/stall transitions at configured fast test cadence. The default production cadence remains slower.